[email protected] +91 9840521421 Scholarly Open Access, Peer-reviewed, Refereed Journal e-ISSN 2583-5300 Since 2022
  1. Home
  2. Archives
  3. 2026
  4. Vol. 5 Issue 1
  5. Article
Original Article

Balanced LSTM-Based Deep Learning Framework for Sentiment Analysis of Amazon Product Reviews

Open Access Peer Reviewed Crossref DOI

Authors

  • Penke Divya1
  • Suneel Kumar Duvvuri2

1 Student, Department of Computer Science, Government College (Autonomous), Rajahmundry, Andhra Pradesh, India. 2 Assistant Professor, Department of Computer Science, Government College (Autonomous), Rajahmundry, Andhra Pradesh, India.

Article Information

Published Online
January-April 2026
Pages
443-451
Article Type
Original Article
Check for updates
check update

Abstract

View PDF

The rapid growth of e-commerce platforms has resulted in an enormous volume of user-generated textual data in the form of customer reviews. These reviews play a crucial role in influencing purchasing decisions and provide valuable insights into product quality and customer satisfaction. However, analyzing such large-scale textual data manually is inefficient and time-consuming. Sentiment Analysis, a subfield of Natural Language Processing (NLP), offers automated techniques to classify opinions expressed in text into categories such as positive, negative, and neutral. This research proposes a robust sentiment analysis framework using a balanced dataset derived from the Amazon Fine Food Reviews corpus. One of the major challenges in sentiment classification is class imbalance, where positive reviews dominate the dataset, leading to biased predictions. To address this issue, a balanced dataset consisting of 150,000 reviews equally distributed across sentiment classes is constructed using an under-sampling approach. The proposed system integrates advanced NLP preprocessing techniques, including tokenization, stop-word removal, and stemming, followed by feature extraction using word embedding’s. A Long Short-Term Memory (LSTM) deep learning model is employed to capture contextual dependencies within textual data. The model is trained and evaluated using standard metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed model achieves a test accuracy of 94.37%, indicating a significant improvement over traditional and imbalanced learning approaches. The use of a balanced dataset effectively reduces model bias and enhances classification reliability across all sentiment classes. The proposed framework can be effectively applied in real-world applications such as customer feedback analysis, product evaluation, and decision support systems.

APC from ₹1200 + 18% GST

Peer-reviewed open access₹1200 Indian (₹1400 with DOI)$80 USD internationalpaid after acceptance Peer-reviewed open access₹1200 Indian$80 USD international ₹1200 Indian · $80 intl

  • CrossRef DOI
  • ISSN 2583-5300
  • Open Access
  • Peer Reviewed
  • Plagiarism Checked
  • Google Scholar
Contact Us Submit Paper